Coexistence Console

A Devvit moderation console for communities where humans and AI create together, with disclosure rules, review queues, transparency labels, and mod-first decision workflows.

Comment

Inspiration

I built Coexistence Console because I am genuinely a fan of Reddit.

Reddit is one of the few places on the internet where people can talk freely, ask strange questions, share deep expertise, argue, learn, vent, and build culture together. If you love Reddit, you eventually start thinking about the people who keep those spaces working: moderators.

The next era of Reddit will not be purely human-written. AI-assisted posts, AI-generated posts, human posts, and posts with no clear disclosure will all exist in the same communities. I do not think the answer is to ban AI from every conversation. I also do not think moderators should be forced to guess, accuse, or trust unreliable AI-detection scores.

The world I expect is more practical: humans use tools to reduce the moderation burden, some moderation work eventually becomes AI-assisted too, and both human moderators and future AI moderation agents need the same thing: clear rules, disclosure norms, review workflows, action history, and accountability.

That is also how this project was made: a human and AI working side by side, checking each other's work, keeping records, and deciding what was safe to ship. That experience made the moderator problem feel concrete. Coexistence is not an abstract idea; it is people and AI doing real work together, with humans still responsible for judgment.

That is why Coexistence Console is not an AI detector. It is a governance console for coexistence.

What It Does

Coexistence Console gives subreddit moderators a complete workflow for AI-era community governance:

  1. Policy Editor Moderators choose AI-friendly, Balanced, Strict, or Custom governance, then tune the workflow for community types like advice/support, technical, creative, education, marketplace, and general discussion.

  2. AI Policy Copilot Gemini helps draft moderator-reviewed policy text, disclosure request copy, removal reasons, sidebar/wiki text, review checklists, and recommended workflow settings. The AI drafts only; a moderator must review and save the result.

  3. Rule-Based Review Queue Real subreddit posts are scanned with deterministic policy signals such as missing disclosure, low context, repetitive structure, similarity to recent posts, and policy mismatch. The queue never says "AI detected" or "confirmed AI."

  4. AI Triage Assistant For a queued post, Gemini can suggest a safe label candidate, two observable reasons, and a next moderator action. If disclosure is unknown, it cannot label the post as AI-assisted or AI-generated. It routes the case toward disclosure or review instead.

  5. Moderator Actions Moderators can approve, apply label, ask disclosure, mark reviewed, or remove with reason. These actions target real Reddit post IDs and are written to ActionLog.

  6. ActionLog-Based Analytics Dashboard and Analytics are built from real ActionLog and Redis state, not fixed demo numbers. They show scanned posts, queue items, labels applied, disclosure requests, removals, approvals, reviewed posts, coexistence visibility, and estimated time saved.

  7. Community Pulse The app summarizes aggregate mood, review pressure, disclosure clarity, and the next best moderator move. Gemini can generate a short moderator-facing pulse summary from aggregate signals.

  8. Multilingual Moderator UI Core moderator surfaces support English, Japanese, Spanish, and French, with an in-console language switcher and translation preview for queued post content.

What This App Does Not Do

Coexistence Console intentionally avoids the dangerous parts of AI moderation:

  • It does not accuse users of being AI.
  • It does not display AI-detection scores.
  • It does not ban users automatically.
  • It does not let AI make removal decisions.
  • It does not profile users across subreddits.
  • It does not train models on Reddit data.

Unknown disclosure is treated as an operational state for review, not proof that someone is deceiving the community.

How We Built It

The app is built entirely as a Devvit moderation app.

  • UI: Devvit Blocks custom posts for Dashboard, Review Queue, Policy Editor, and Analytics.
  • Triggers: PostSubmit and delete-related triggers keep review state and cleanup in sync.
  • Storage: Devvit Redis stores subreddit policy, review items, action logs, recent snippets, metrics, and language preferences.
  • Signals: Policy matching, low-context checks, repetitive structure checks, similarity signals, and disclosure-state handling are deterministic.
  • AI usage: Gemini is used only for moderator-assist surfaces: policy drafting, AI Triage suggestions, translation preview, and aggregate pulse summaries.
  • Verification: The final QA release candidate was tested live on Reddit playtest with build, dashboard, queue, Japanese UI, and live AI verification scripts.

Final verified app version: super-consolex v0.0.54.

Challenges We Ran Into

The hardest product decision was not technical. It was deciding what not to build.

An obvious version of this idea would have been an AI detector. But that would be unfair, brittle, and not very useful for moderators. The more useful product is a workflow that gives moderators visibility without turning uncertainty into accusation.

On the technical side, the main challenges were:

  • Keeping Devvit trigger data, Redis state, and custom post UI in sync.
  • Making queue actions affect real Reddit posts and real ActionLog rows.
  • Making analytics derive from stored actions instead of fake dashboard values.
  • Keeping AI usage safe, moderator-facing, and optional.
  • Handling multilingual UI state reliably inside Reddit.
  • Hardening Playwright verification against Reddit navigation and Devvit hydration timing.

Accomplishments That We Are Proud Of

  • The product works as a real Devvit app, not a mock dashboard.
  • Real subreddit posts can enter the queue.
  • Moderator actions can operate on real post IDs.
  • ActionLog drives analytics and estimated time saved.
  • AI Policy Copilot, AI Triage Assistant, translation preview, and AI Pulse all work as moderator-assist features.
  • The app keeps the safety boundary clear: signals, not accusations; human control, not automatic punishment.
  • The UI is polished enough to explain the workflow quickly to a moderator.
  • The core experience supports English, Japanese, Spanish, and French.

What We Learned

Moderation in the AI era is not just about detecting AI. It is about helping communities decide how AI can participate, how people should disclose it, and how moderators can stay consistent without becoming overloaded.

I also learned that the best place for AI in a moderation tool is not necessarily enforcement. It is often setup, summarization, drafting, translation, triage, and explanation, with the moderator still in control.

Project Impact

Coexistence Console is designed for communities where AI use is already becoming normal:

  • Advice and support communities can ask for disclosure before taking harsher action.
  • Technical communities can allow AI-assisted explanations while reviewing low-context or repetitive answers.
  • Creative communities can label AI-generated work transparently.
  • Education communities can support learning while making AI assistance visible.
  • Marketplace and review communities can reduce low-context or templated promotional content.

The measurable impact is moderator time saved and decision consistency:

  • Queue items are grouped by explainable reasons.
  • One-click actions reduce repeated manual work.
  • ActionLog creates a shared record for mod teams.
  • Analytics show scanned posts, labels, requests, removals, approvals, reviewed posts, and estimated time saved.

Estimated time saved is intentionally simple and transparent:

estimated time saved = one-click moderator actions x 30 seconds

Why This Matters For Reddit

Reddit communities are different from generic social feeds. They have culture, norms, inside jokes, expertise, and volunteer moderators who understand the local context.

As AI participation increases, the best outcome is not a blunt AI ban or a fake sense of certainty from detection scores. The better outcome is governance: communities decide their own norms, moderators get better tools, and users understand what is expected.

Coexistence Console is built for that future.

It helps today's human moderators. And if moderation becomes more AI-assisted in the future, those AI moderation agents will need the same rails too: policy, disclosure, review reasons, action logs, analytics, and accountability.

What's Next

  • App Directory launch hardening
  • Mobile layout testing
  • Subreddit rule import for faster setup
  • Moderator feedback loops for tuning policy defaults
  • More community-type presets
  • Wider language coverage

Built With

  • devvit
Share this project:

Updates